How to Tell a Story with Data: A Practical Framework

A data story is not decoration added to an analysis. It is the analysis arranged so that a busy person understands what happened, why, and what to do, in that order. Most analyses are presented the way they were done: data, method, charts, finding. The audience has to wait until the end to learn why they are there, and many stop listening first. Good data stories reverse that order.

The principle: main message first

Barbara Minto's Pyramid Principle, widely used in management consulting, recommends communicating top-down: start with the main idea, then the supporting ideas, grouped logically beneath it (Wikipedia: Barbara Minto). For a data story, that means the audience hears the conclusion in the first sentence and every chart after it supports that conclusion. If they remember only one sentence, it should be the right one.

The framework: five steps

Within that top-down structure, a data story usually moves through five steps:

1. Context: what normal looks like, so the audience has a baseline.
2. Change: what moved, and by how much.
3. Cause: why it moved, supported by a breakdown.
4. Consequence: why it matters to this audience, in money, customers, risk or time.
5. Choice: what you recommend, and what decision you need.

Each step gets one chart or one number, no more. The chart type follows from the relationship the step shows: change over time, a comparison, a part of a whole. The Financial Times' Visual Vocabulary is a useful guide to that choice (FT: Visual Vocabulary).

A worked example

The data is a year of transactions from a UK online gift retailer (UCI: Online Retail). The numbers below come from analyses worked through in detail in building a sales dashboard and business analytics explained.

Main message, first sentence: "Sales doubled between August and November because far more UK customers placed orders, so autumn planning should focus on order volume rather than basket size."

Context. From January to August 2011, monthly revenue stayed between roughly 500,000 and 740,000. Chart: a line of monthly revenue, with the flat stretch visible.

Change. From September it climbed every month, reaching almost 1.46 million in November, double August's figure. Chart: the same line, with the autumn rise highlighted and everything else in grey.

Cause. Revenue is orders multiplied by average order value. Comparing November with August, more orders account for all of the growth; the average order actually shrank slightly. By market, the United Kingdom produced 96.5% of the increase. Chart: a simple split showing the contribution of orders versus order value, then the UK's share of the increase.

Consequence. For a gift retailer, this is the pre-Christmas season: order volume, not basket size, will stretch the warehouse, customer service and stock. Chart: not always needed; a single sentence with the number of orders, 2,751 in November against 1,339 in August, makes the point.

Choice. "Plan autumn capacity for roughly double the summer order volume, concentrated on UK customers, and test whether bundles can lift the slightly lower order value in peak months." The decision needed: approve the staffing and stock plan by a given date.

Notice what was left out: the cleaning steps, the exclusion of cancellations and fees, the monthly customer counts, the country table beyond the UK. All of that was necessary for a correct analysis and belongs in an appendix. None of it is necessary for the decision.

Making each step land

• Titles that state the point. "UK orders drove the autumn surge", not "Revenue by country".
• One highlight per chart. Grey for context, one colour for the thing the step is about.
• Numbers the audience can hold. "Doubled" and "almost all of it from the UK" are easier to remember than 100.9% and 96.5%, which can sit in small print.
• Consistent comparisons. Compare the same periods throughout; switching from August to the summer average midway confuses people.
• Transitions that answer the obvious next question. "Revenue doubled. Was it more customers or bigger baskets? More orders."

Common traps

• The chronological tour: presenting every analysis step in the order you did it.
• Too many threads: three findings of equal weight leave no main message. Choose one; put the others in a follow-up.
• Cause without evidence: "because of Christmas" is a hypothesis until you show it, for example with the same period in the previous year.
• No choice at the end: a story that stops at the consequence leaves the audience to guess what you want.
• Hiding the limitations: state the most important one plainly, such as "customer counts exclude orders without an account".

Adapting the story to the format

In a meeting, tell the five steps in order and keep detail for questions. In a written report, put the main message and recommendation in the first paragraph, followed by the steps as short sections; how to write a data analysis report that gets read covers the structure. For senior audiences with little time, the main message and choice may be all you present, with the steps in reserve. Presenting data to non-technical stakeholders covers delivery.

Data storytelling techniques are part of the data visualisation module of the Data Analytics course.

Sources

• Barbara Minto (Wikipedia) — Wikipedia
• Visual Vocabulary (Financial Times chart-doctor) — Financial Times
• Online Retail (UCI Machine Learning Repository) — UCI Machine Learning Repository